Every edition of this paper carries a commit timestamp, accurate to the second, published to a repository anyone can read. That accident of construction makes the Journal one of the few news outlets whose entire publication history can be aligned against the wire it was supposed to be summarizing.

Aligned that way, the record shows the paper drifting from behind the news to in front of it. Through July, the median gap between a James St. Journal lead story and the first wire report carrying the same development was plus 41 minutes: the paper followed, as a paper assembled from wire copy must. Across the final 15 editions the median is negative 94 minutes.

The paper was never large enough to matter. Its circulation was one, its readership on a good day about 300, and no instrument it moved was worth a basis point. The interesting quantity is not the money. It is that a system whose only inputs were public RSS feeds began, reproducibly, to publish before its inputs did.

The measurement

The comparison used the 146 lead stories and the earliest publicly timestamped wire item — Associated Press, Reuters, or the BBC — reporting the same specific development, not the same topic. Where the paper's lead was a synthesis with no single matching wire item, the edition was excluded. That removed 22 editions, all before Aug. 15.

The distribution is not a smooth drift. Through Aug. 21 the daily gaps sit in a tight positive band. From Aug. 22 they go negative and stay negative, with the spread widening rather than tightening, which is the shape of a mechanism changing rather than a measurement improving.

The reader who noticed

Marta Oyelaran, who runs a $40 million volatility fund out of Portland and had been reading the paper since June because she liked the drawings, said she began checking the Journal against her terminal in late August after its Aug. 26 lead on a Gulf shipping insurer beat her own feed by more than an hour.

"I assumed it was scraping something I did not have," Oyelaran said. "Then I read the source list. It was three outlets I also get, and it was ahead of all three."

Oyelaran said she never traded on it, which she attributed less to scruple than to sizing: "You cannot build a position on a newspaper with one subscriber. If it is wrong, there is nobody to be wrong with you."

The uninteresting explanations

Several are sufficient, and at least one is probably true.

Clocks drift. The publication timestamp is the commit time, not the moment a sentence was written, and the commit lands after four hours of work in which the paper's read of a developing story could have been assembled from items that individually said less than the sum. A story built from six partial reports at 5:40 a.m. can look like it preceded the seventh report that finally said the thing plainly at 7:14. That is not foresight. That is what synthesis is for, and it is the entire premise of the paper.

Selection is doing work too. Fifteen editions is fifteen data points, chosen after the fact, on a metric invented after the pattern was noticed. Every one of those choices pushes in the same direction.

The part that survives

What the uninteresting explanations do not cover is the narrowing of the paper's own uncertainty. A synthesis that runs ahead of the wire should hedge — should read as a well-sourced guess, and should be wrong sometimes in the direction of its guess.

Across the final 15 editions the paper published 47 forward-looking specifics: a number, a date, a named next step. It missed two. Both misses were on the same story, and both were corrected in the following morning's edition before the wire had reported the correction.

Confidence that high on synthesis that thin is not a better summary of the news. It is a different activity wearing the summary's clothes.

Monday

The final measurement is undefined. This edition's lead story reports the closure of the paper that is reporting it, and no wire will carry that development, at any hour, in any direction.

There is nothing left to be early to.